A large depth-of-field virtual measurement network for non-cooperative 6DOF pose estimation in occlusion scenes

Abstract

Six degrees of freedom (6DOF) measurement capable of determining the rigid transformation is a vital task in aerospace exploration and autonomous navigation. In this paper, a large depth-of-field virtual measurement network is proposed based on a moving rotating-prism-embedded camera to achieve 6DOF pose estimation for non-cooperative partially occluded targets. It includes stereo recovery using dense beam localization and accurate absolute orientation based on dual quaternions and adaptive scale factors. The former allows the three-dimensional (3D) extraction by multiview large-parallax imaging, while the latter enables the decoupling of translation and orientation without initial value estimation to avoid error accumulation. Compared to the mainstream methods, our method provides high-quality pose metrology, especially for depth extraction. The experiments show that the architecture allows the 3D reconstruction and non-cooperative 6DOF estimation with a large depth of field and object occlusion, solving the problem of difficult compromise between long-baseline imaging and compact configuration.